GROMACS上如何进行GPU加速.pdf

Molecular Simulation with GROMACS on CUDA GPUs Erik Lindahl WebinarWe’re comfortably on the single-μs scale today Larger machines often mean larger systems, not necessarily longer simulations GROMACS is used on a wide range of resources Why use GPUs? Throughput Performance ? Sampling ? Free energy ? Cost efficiency ? Power efficiency ? Desktop simulation ? Upgrade old machines ? Low-end clusters ? Longer simulations ? Parallel GPU simulation using In#niband ? High-end efficiency by using fewer nodes ? Reach timescales not possible with CPUs Caveat emperor: It is much easier to get a reference problem/algorithm to scale i.e., you see much better relative scaling before introducing any optimization on the CPU side When comparing programs: What matters is absolute performance (ns/day), not the relative speedup! Many GPU programs today Gromacs-4.5 with OpenMM Gromacs running entirely on CPU as a fancy interface Actual simulation running entirely on GPU using OpenMM kernels Only a few select algorithms worked Multi-CPU sometimes beat GPU performance... Previous version - what was the limitation? Why don’t we use the CPU too? ~2 TFLOP0.5-1 TFLOP Random memory access OK (not great) Random memory access won’t work Great for throughput Great for complex latency-sensitive stuff (domain decomposition, etc.) Programming model CPU (PME) GPU N OpenMP threads 1 MPI rank 1 MPI rank 1 MPI rank 1 MPI rank N OpenMP threads N OpenMP threads N OpenMP threads 1 GPU context 1 GPU context 1 GPU context 1 GPU context Domain decomposition dynamic load balancing Load balancingLoad balancing Gromacs-4.6 ext-generation GPU implementation: Heterogeneous CPU-GPU acceleration in GROMACS-4.6 Wallclock time for an MD step: ~0.5 ms if we want to simulate 1μs/day We cannot afford to lose all previous acceleration tricks! ? Δt limited by fast motions - 1fs ? Remove bond vibrations ? SHAKE (iterative, slow) - 2fs ? Problematic in parallel (won’t work) ? Compromise: constrain h